Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

Explainable Coarse-to-Fine Ancient Manuscript Duplicates Discovery

Created by
  • Haebom

Author

Chongsheng Zhang, Shuwen Wu, Yingqi Chen, Yi Men, Gaojuan Fan, Matthias A{\ss}enmacher, Christian Heumann, Jo ao Gama

Outline

This paper presents a novel framework for identifying duplicates in ancient documents. Specifically, we design an advanced duplicate detection framework that combines low-level keypoint matching and high-level text-centric content-based matching for one of the ancient documents, Oracle Bones (OB). Compared with existing content-based image retrieval and image matching methods, our model achieves similar recall performance and higher simplified average inverse rank score, while achieving much faster computational efficiency. Through practical applications, we have discovered more than 60 new pairs of OB duplicates that traditional experts have failed to detect for decades. We have released the code, model, and actual results on GitHub.

Takeaways, Limitations

•
Takeaways:
◦
A new efficient and accurate framework for identifying duplicates in ancient documents is presented.
◦
Demonstrating the utility of an approach that combines low-level and high-level features.
◦
Discovering new duplicates and contributing to archaeological and historical research through real-world applications.
◦
Achieve improved performance (recall rate, average inverse rank score, computational efficiency) over existing methods.
•
Limitations:
◦
Since this model is specific to Oracle Bones, its applicability to other types of ancient documents may be limited.
◦
Further research is needed on the model's generalization performance and robustness to different types of redundancy (e.g., partial redundancy).
◦
Additional expert review is needed to verify the accuracy of duplicate determinations.
👍
Made with Slashpage